I would like to know what is the percentage of amino acids that has to
lie in the favoured region (only favoured region) in ramachandran plot
after modeling the protein.
I wouldn't say there's a strict cutoff above which your model should be
considered "bad". However, Modeller's restraints are designed to place
dihedrals in the favored regions, and our assessment scores should
penalize models that have residues in disfavored regions.